*4.1.2 Lag length criteria*

about the magnitude and duration of the shock proportion of a variable to the variable itself and to other variables. According to Basuki [26], variance decomposition aims to measure the magnitude of the contribution or composition of the

According to Basuki [26], the Granger causality test is used to see whether two

The VECM estimation is started by testing the data stationarity of each variable as the initial process. To detect the stationarity of each variable, the ADF test is used with the intercept model. Data sets are declared stationary if the average values and variants of the time series data do not change systematically over time or the averages and their variants are constant [29]. The ADF stationary test for each

According to **Table 1**, at the level, there is no single variable that meets stationary requirements, either from FDR, NPF, or BOPO. It is indicated by the value of t-ADF which is greater than the Mackinnon critical value, so it is necessary to test at

variables have a reciprocal relationship or not. The variable can have a causal relationship with other variables significantly. It implies each variable has the

influence of each independent variable on the dependent variable.

opportunity to become an endogenous or exogenous variable.

*3.5.8 Testing Granger causality*

*Vector error correction model. Source: Gujarati [28].*

**Figure 2.**

*Banking and Finance*

**4. Result and analysis**

*4.1.1 Unit root test*

**58**

**4.1 Causality test and data instruments**

variable can be indicated as follows.

the first difference level shown in **Table 2**.

The lag length is used to determine the effect of the time taken from each variable on the past variable. The selected lag candidates are the length of lag according to the *likelihood ratio* (LR) criterion, *final prediction error* (PPE), *Akaike information criterion* (AIC), *Schwarz information criterion* (SIC), and *Hannan-Quinn criterion* (HQC). The determination of the optimal lag length in this study is based


#### **Table 1.**

*Unit root test-augmented Dickey-Fuller (level).*


#### **Table 2.**

*Unit root test-augmented Dickey-Fuller (first difference).*

### *Banking and Finance*

on the sequential modified LR test statistical criteria. The lag length that was included in this study is from 0 to 3.

Based on **Table 3**, the optimal lag on all variables from FDR, NPF, and BOPO is in lag 3, that is, with the sequential modified LR test statistic 24.77971, PPE 4.037246, and AIC 9.907182. Therefore, the optimal lag has been statistically determined and the VAR stability test is carried out.
